Recent advancements in Earth Observation (EO) foundation models and their embeddings—such as Prithvi and AlphaEarth—provide unprecedented opportunities for biodiversity conservation. These models and embeddings offer the potential to unify disparate geospatial datasets for species distribution modeling, habitat characterization, and ecosystem health assessment. However, a significant gap remains in translating these high-capacity computational architectures into robust ecological tools. This session examines research where ecologists and conservation researchers integrate EO foundation models with biodiversity data to address ecological questions and monitoring needs. We focus on novel methods for fine-tuning models on taxonomic observations, quantifying spatial-temporal biases, and understanding the applicability of embeddings to different tasks. By highlighting emerging research avenues and methodological best practices, this session aims to define the current state of EO-AI in biodiversity science. Discussions will center on overcoming technical hurdles, such as input requirements and model interpretability, to provide actionable insights for large-scale conservation monitoring and ecological decision-making.

• IN - Informatics